# SpaceApp: every tool > Every tool of SpaceApp, from the room you work in to the mind that runs it, one section each. Reviewed: 2026-10-27. The tools ship with the public installer from run-spaceapp 1.1.9; the release notes at https://spaceapp.dev/docs/changelog/ list what each version brings. Index: https://spaceapp.dev/llms.txt Installation: https://spaceapp.dev/llms-install.txt AI agents: https://spaceapp.dev/llms-agents.txt Architecture: https://spaceapp.dev/llms-architecture.txt ## 1. Every tool at a glance SpaceApp is one self-hosted workspace for your AI coding agents and the work around them. The tools fall into six groups. | Group | Tools | Read more | | --- | --- | --- | | Workspace | Rooms and panes, 16 AI coding CLIs, the Room Agent, managed browsers, files and Agent Files, scheduled tasks, team rooms | [Rooms & panes](/docs/rooms-and-panes/), [AI coding CLIs](/docs/agents/) | | Nous | Nous, Needs you, the Plan map | [Nous](/docs/nous/), [Needs you & Plan map](/docs/needs-you/) | | Memory and research | Space Memory, the Research Council, Live voice | [Space Memory](/docs/memory/), [Research Council](/docs/council/), [Live voice](/docs/voice/) | | Logins and model accounts | Vault, Space Gateway, token usage and your bill | [Vault](/docs/vault/), [Space Gateway](/docs/gateway/) | | Copilots and studios | Mail, Job, Marketing, Freelance, Security and Monitoring Copilots; the Design, Video and App Studio panes; Media Studio | [Copilots](/docs/copilots/), [Studios](/docs/studios/) | | Devices | Any browser on a computer, phone or tablet; the Android app; desktop builds | [Devices](/docs/devices/) | Every tool runs on your own host and on your own AI accounts. Start with one room and OpenCode, and turn the rest on as your work grows. ## 2. Rooms and panes A room holds one piece of work. Its panes sit side by side: up to 16 in a room, each a CLI, a browser, a copilot, a studio or another tool. There are 35 pane types in the Create menu, in four groups: 16 AI coding CLIs, 6 copilots, 3 studios and 10 more panes (Browser, Live, Files, Databases, YouTube, VNC, Chat, DeepSeek Harness, T3 Code and Demo Projects). Rooms keep their sessions while you switch between them: up to six rooms stay warm, so a CLI keeps its process, scrollback and conversation when you come back. A pane can be maximized, minimized to the bar or floated; layouts sort themselves or follow the arrangement you choose, and each room can take one of six themes. Every pane in a room shares the room's workspace: the project folders you registered, the same skills, the approved MCP servers and Space Memory. ## 3. AI coding CLIs Sixteen coding CLIs run in panes, each with its own branding and controls: OpenCode, Codex, Claude Code, Gemini, Autohand Code, Qwen Code, Kimi Code, Grok Build, DeepSeek, Cursor, GitHub Copilot, Hermes, Oh My Pi, Qoder, Muse Code and Droid. OpenCode comes set up with SpaceApp and starts on a free model, so the first agent needs no subscription or key. Ten more install with one launcher command each (`provider install `), and every CLI signs in with your own account or key. A pane's account menu moves a CLI between your direct account and Space Gateway, and its model menu changes the model of that session only. ## 4. The Room Agent Each room has an agent that watches every pane in it and shows which ones wait for you. Talk to it about the room; with Auto-answer on it answers the prompts that wait, so the agents keep working while you are away. It reads which files each pane's worktree changes and warns when two panes change the same files, and their work joins one release train instead of two competing versions. When a CLI's turn breaks on a server error, it types "continue" at the pane's empty prompt a few times an hour, never into a draft. ## 5. Managed browsers Browser panes are live Chrome sessions on your host, with tabs, device viewports and your Google accounts. Agents drive them with their own browser tools; when a page needs a person (a sign-in, a CAPTCHA, a code), the agent hands it to you with one request and takes it back when you are done. One list of bookmarks serves every browser pane and copilot, and each bookmark keeps the Google account it was saved with. Browser panes need the large resource profile. ## 6. Files, clipboard and Agent Files The Files pane browses your registered project folders. The clipboard dock keeps your private copy history, notes and plans; a plan carries its steps and progress, and the pane working on it reports each finished step. Agent Files is the dock for finished work: the files you asked an agent to make, sorted into folders, with a preview, details and download. The Media dock keeps the screenshots, videos and other media of your rooms. ## 7. Scheduled tasks The Tasks dock is one list of every task with its time, yours and your agents', as a list or a month calendar. A task can repeat: once, daily, weekly or every N minutes. When a task is due, SpaceApp types it into its CLI pane, only at an empty prompt. A busy pane waits, a closed pane reopens on its conversation, and a missed occurrence is recorded as skipped. A task without a pane is a reminder. Agents add timed work there too, in your working time zone, instead of a cron job that dies with the session. ## 8. Team rooms Each person on a team gets an account and a CLI container of their own, so nobody reaches another person's keys or repositories. The owner of a room can share it with a teammate to view it or to work in it; each pane runs in the container of the person who opened it, and only its owner types into it. Memory stays personal. ## 9. Nous One mind for your whole space. Nous sees every room, pane, copilot, plan and release train, and reads each pane's progress reports instead of its screen, so it knows what every agent does and what it waits on. You talk to Nous instead of to every agent. It does the routine work itself: it opens, steers and closes panes, sends them their next step, runs up to five helpers side by side, saves finished work to Space Memory and closes panes whose work is done. Standing orders run around the clock: finished panes are saved and closed, stalled plans get a nudge, and a daily self-review looks back at its own turns. It brings you only four kinds of decisions: releases, deletes, anything that reaches third parties or money, and the direction of the work. Each decision is a card with Approve and Decline; a plain yes or no in words decides it too. "Always approve" lets a routine kind of decision run on its own next time, never a release or a delete. You reach it in its dock, on your phone or in a paired Telegram chat. While you are away it keeps a log for your return; at night, requests for your hands wait for the morning; and it spends less while your AI quotas run low, with a cheaper model for its background turns. ## 10. Needs you One attention request for every agent, pane, copilot and Nous. When an agent wants your hands or an answer, it asks one question with its options and a recommendation. A banner names the pane and goes there with you, the pane opens maximized on what needs you, and a Back pill returns you to where you were. You can also answer from the Needs you list in the Nous dock: pick an option or type a reply, and the answer goes to the pane, the copilot or Nous that asked. A request no screen showed stays open until you see it. ## 11. The Plan map One picture of every open plan: what each one waits on (you, a date, a promote, a third party, Nous or the server), the plans it depends on and the date it must keep. "Unblock first" lists what someone can act on, with the number of plans each blocker frees; the dates form one timeline. Nous reads the same map before it proposes or starts work. ## 12. Space Memory Decisions and project knowledge kept as records that every agent session reads and writes. A decision made in one session on Monday is there for a new session in another CLI on Tuesday: "use the approach we agreed on" finds it. Records carry tags, a source and dates, and newer decisions supersede older ones. The app shows them as a list and as a graph, with keyword and vector search. Embeddings run on your server's own CPU with a local model, so memory text stays on your host. Agents save with `space-memory-save` and recall with `space-memory-recall`; secrets never belong in memory (the Vault keeps them). ## 13. The Research Council Several models work on one question. A coordinator plans the run from your question, your profile and your memory, asks you what is missing and waits for your approval of what is shared about you. The members answer on their own from the same web evidence pack; the answers are merged into positions without names; every member scores every position; the coordinator writes the agreed Top 3. A finished run becomes a plan in your clipboard with one click. ## 14. Live voice A real conversation with your workspace. Live opens rooms, starts agents, directs panes and tells you what is done, and it sees your screen when you share it. It runs on Gemini Live or on Fish Audio's Drama 3 voice, in English, Greek and other languages. ## 15. Vault A password store your agents can use but not read. Your browser encrypts every login with keys that live only in the open tab; the database keeps ciphertext, public keys and metadata. A root-owned broker on your host fills a login or a one-time code into the pane's browser, and the agent gets back a status and an audit id, not the value. After a fill the page stays protected until it is gone: the agent can click, type and press keys but cannot read the field. One switch turns agent access on or off, and logins import from a Chrome password export. It also keeps server passwords and API keys: an agent asks for one by its id, and the value is written straight into the file that needs it, not into the agent's context. Logins that agents create, and passwords that reach them in a mail, are saved to the Vault rather than to memory or notes. ## 16. Space Gateway One gateway on your host for your model accounts: Claude, Codex and Gemini subscriptions and an Anthropic API key. Calls go to the subscriptions first and to the API key last, and return to the subscriptions after their limits reset. Each tool gets its own key with its own models and daily limits. CLIs, copilots and Nous use it the same way, and a CLI pane can move to it when its direct account reaches its usage limit. ## 17. Token usage and your bill Every token is counted once, per model, provider and account, with its API value at list prices: the gateway's own request logs plus the CLI calls that did not go through it. With your Anthropic Admin key, the gateway reads Anthropic's real bill every few minutes and shows today, 7 days, this month and 31 days next to your monthly budget and the credits you entered. ## 18. Copilots: how they work A copilot is a pane for the work around the code, with its own CLI agent, memory, rules and activity feed. You answer its questions and reviews in the rail and choose how much it does alone: Assist drafts and asks, Rules acts only under rules you approved, Full acts on its own. Every send, publication or change passes a check on the server first. Copilots are optional modules: new installs start with none of them. Turn one on with `optionalModules` in the launcher's `config.json`, then run `npx --yes run-spaceapp@latest repair`. Copilots that use the pane browser need the large profile. ## 19. Mail Copilot A Gmail client for your connected Google accounts, with an agent that works through your mail. Under Assist it drafts replies for your review; under Rules it acts within rules you approved; under Full it acts alone. It learns from what you do with mail, on the web and on your phone too: a consistent habit becomes a rule it proposes. "Send to AI" hands it a mail to check and fix what it needs. ## 20. Job Copilot Works through a queue of job postings in the pane browser: it fills applications from your CV and profile, researches the salary of each posting and recommends one, asks you for what it cannot answer, hands you CAPTCHAs, and keeps a tracker of every application with evidence. ## 21. Marketing Copilot Drafts posts from your briefs and campaigns and publishes them on your own accounts through the pane browser: X, LinkedIn, Reddit, Hacker News, Product Hunt, dev.to, YouTube and GitHub. Each channel has its own autonomy cap and post limit; facts come from the public product knowledge, and the published URL is the evidence. It measures each post a day and a week later. ## 22. Freelance Copilot Finds client work on Upwork, Contra, Fiverr, Toptal, Arc and Braintrust, qualifies leads and writes proposals and messages under daily caps and a budget. Each send is checked for wording and scam signals first, and a client who wants to go ahead is handed to you. ## 23. Security Copilot Watches CISA KEV, NVD, OSV, distro advisories and security news, and matches them to the packages and container images of your hosts with a fixed read-only probe and Trivy. It fixes what a finding needs through a gate: a backup first, a check after, a rollback ready; reboots, deletes and payments wait for a person. Fixes to code go through your normal review. ## 24. Monitoring Copilot Reads your Prometheus, Alertmanager and Loki collector and explains each alert in plain words. A critical alert reaches Nous; a warning is noted; a resolved alert closes its card. It repairs what a playbook covers through the same kind of gate: restarts and config changes with a backup, reboots and deletes only with your yes. ## 25. Design pane A graphics studio any agent or pane can call. It makes images with Nano Banana or gpt-image through Space Gateway, sets them in HTML templates and brand kits at the exact size of every social format, adds motion, and delivers them to the agent or pane that asked, at once or after your approval. Every image carries a description and who made it. ## 26. Video pane A video studio with the same idea: send it a video or a brief and get back finished videos in every format you ask for. It cuts by silence, scene or highlights from the transcript, adds transitions, filters, captions, narration and music with its credits, and delivers each format with a poster frame. A YouTube video's own captions give the full text of what was said. ## 27. App Studio Builds mobile apps for Android and iOS with Expo, previews every screen on ten current phones as images from the app's web export, and builds a test APK signed with a debug key. Nous and you send it briefs; it asks you what only you can decide. ## 28. Media Studio Screenshots and short videos of your own workspace in device and social sizes (16:9, square, portrait, story), with framed social cards and private text masked. Captures of your live rooms are read only. ## 29. Any device One install on your own host, and the same rooms, panes and agents from a browser on a computer, phone or tablet. The host must stay on and be reachable from your other devices; see [Remote & mobile access](/docs/remote-access/). ## 30. The Android app Signs in with your Google account through the system browser and keeps a device session you can revoke in Settings. SpaceApp notifications, Needs you requests and Nous cards reach your phone, and a tap opens the pane that needs you. ## 31. Desktop apps Windows and Linux builds of SpaceApp that open your workspace in its own window, with desktop notifications and an update notice.